The III-V compound semiconductor Gallium Nitride (GaN) and its Aluminum (Al) and Indium (In) alloys are ideal materials both for high-frequency and high-power electronic applications (see for example Brown et al., Solid-State El. 46, 1535 (2002), the content of which is incorporated herein by reference thereto). These materials are also ideal for short wavelength light emitting diodes and lasers (see for example Nakamura, Annu. Rev. Mater. Sci. 28, 125 (1998); Nakamura, Science 281, 956 (1998); and Smith et al., J. Appl. Phys. 95, 8247 (2004), the contents of which are incorporated herein by reference thereto).
One of the main drawbacks of the material is, however, the lack of large, single crystals due to extreme conditions of high temperature and pressure required for their growth in bulk form. The only way to synthesize GaN wafers of significant size is by means of heteroepitaxy, whereby thick, self-supporting GaN layers are grown onto a support substrate such as sapphire or Silicon Carbide (SiC), which substrate is subsequently removed. Thinner heteroepitaxial III-V nitride layers can be used for device processing without removal of the substrate.
One common problem of all techniques used for heteroepitaxial growth of GaN is the high dislocation density initially present in the growing layers. This problem results from the different lattice parameters of GaN and the available substrate materials, such as sapphire, silicon carbide and silicon (see for example Dadgar et al., Phys. Stat. Sol. (c) 0, 1583 (2003), the content of which is incorporated by reference thereto). As a consequence of the high misfit dislocation density, heteroepitaxial GaN layers tend to contain also a high density of threading dislocations (TD), which degrade device performance whenever they penetrate into any active layers. Many ways have been devised to reduce TD densities to values acceptable for device fabrication, such as buffer layer growth of various forms, or lateral overgrowth with and without the use of masks (see for example Davis et al., Proc. IEEE 90, 993 (2002), the content of which is incorporated by reference thereto).
The main methods used for growing epitaxial HI-V nitride layers are hydride vapor phase epitaxy (HVPE), metal-organic chemical vapor deposition (MOCVD) and molecular beam epitaxy (MBE). In HVPE, pure metals are used as source materials, and transported as gaseous halides to the reaction zone where they react with a nitrogen-containing gas, usually NH3 to form an epitaxial layer on a substrate typically heated to above 1000° C. HVPE has the advantage of very high growth rates of up to 100 m/h (see for example U.S. Pat. No. 6,472,300 to Nikolaev et al., the content of which is incorporated herein by reference). Because of its high growth rates, HVPE is mostly used for growing layers many tens of microns thick, and in particular for the fabrication of self-supporting layers as substrates for subsequent MOCVD or MBE steps.
Low rates and control of sharp interfaces are, however, more difficult to achieve by HVPE, and may require mechanical movement of the substrate between different reactor zones (see for example U.S. Pat. No. 6,706,119 to Tsvetkov et ah, the content of which is incorporated herein by reference). Additionally, the presence of hydrogen gas in the reaction zone requires annealing of the substrate in an inert gas atmosphere, particularly when high p-type doping for example by Mg impurities is to be attained (see for example U.S. Pat. No. 6,472,300 to Nikolaev et al., the content of which is incorporated herein by reference).
MOCVD (or MOVPE, for “metal-organic vapor phase epitaxy”) is a CVD technique in which metal-organic precursors are used along with other reactive gases containing the anions, such as ammonia in the case of nitride growth. The need for expensive precursor gases, along with rather low growth rates of just a few μm/h, is a significant disadvantage of MOCVD. Furthermore, a buffer layer usually has to be grown for GaN heteroepitaxy on sapphire, SiC or Si, at a lower substrate temperature before the active layer stacks are deposited at temperatures above 1000° C. (see for example U.S. Pat. No. 6,818,061 to Peczalski et al. the content of which is incorporated herein by reference). MOCVD is, however, the technique most often used for growing active layer structures suitable for device fabrication (see for example Wang et al., Appl. Phys. Lett. 74, 3531 (1999) and Nakamura, Science 281, 956 (1998) the contents of which are incorporated herein by reference).
Large differences in thermal expansion coefficients between common substrates and GaN, together with the high substrate temperatures during growth, present a significant obstacle towards achieving crack-free epitaxial layers. Crack avoidance seems to necessitate rather complicated interlayer schemes (see for example Biasing et al., Appl. Phys. Lett. 81, 2722 (2002) the content of which is incorporated herein by reference). HVPE and MOCVD are both deposition techniques working at atmospheric or somewhat reduced pressures. Reactor geometries and gas flows determine layer uniformities to a large extent.
By contrast, in MBE, pressures are in the high-vacuum to ultrahigh-vacuum range and mean-free paths therefore greatly exceed reactor dimensions. Metals are evaporated in so-called effusion cells from which molecular or atomic beams travel towards the heated substrate without being scattered in the gas phase. For nitride growth, a nitrogen source yielding activated nitrogen must be used. Activation is usually achieved by means of plasma excitation of molecular nitrogen. A system for epitaxially growing Gallium Nitride layers with a remote electron-cyclotron-resonance (ECR) plasma source for nitrogen activation has been described for example in U.S. Pat. No. 5,633,192 to Moustakas et al., the content of which is incorporated herein by reference. Since Gallium (Ga) is usually supplied from an effusion cell, MBE does not require the expensive metal-organic precursors common in MOCVD. MBE offers moreover excellent control over layer composition and interface abruptness (see for example Elsass et al, Jpn. J. Appl. Phys. 39, L1 023 (2000), the content of which is incorporated herein by reference). Due to its low growth rates on the order of 1 μm/h and complex equipment, however, it is not considered to be a technique suitable for large scale production of semiconductor heterostructures.
A further method potentially suitable for large scale production of nitride semiconductors (see, for example, U.S. Pat. No. 6,454,855 to von Känel et al. the content of which is incorporated herein by reference) is low-energy plasma enhanced chemical vapor deposition (LEPECVD). In contrast to plasma assisted MBE, where nitrogen activation occurs in a remote plasma source, a dense low-energy plasma is in direct contact with the substrate surface in LEPECVD. The low-energy plasma is generated by a DC are discharge by means of which metalorganic precursors and nitrogen are activated (see, for example, U.S. Pat. No. 6,918,352 to von Känel et al. the content of which is incorporated herein by reference). Potentially, LEPECVD can reach growth rates comparable to those of HVPE (several tens of μm/h, while offering optimum control over the dynamic range of growth rates, such that excellent interface quality can be achieved. Moreover, since activation of the reactive precursors is achieved by means of a plasma rather than thermally, the process is expected to work at lower substrate temperatures. The DC plasma source used in LEPECVD has been shown to be scalable to 300 mm substrates (see for example WO 2006/000846 to von Känel et al., the content of which is incorporated herein by reference).
Although the term “LEPECVD” has been coined in conjunction with a DC arc discharge (see, Rosenblad et al, J. Vac. Sci. Technol. A 16, 2785 (1998), the content of which is incorporated herein by reference), such a DC arc discharge is not the only way to generate a low-energy plasma suitable for epitaxy. According to prior art, sufficiently low-energy ions suitable for epitaxial growth may result also from electron-cyclotron-resonance (ECR) plasma sources (see Heung-Sik Tae et al., Appl. Phys. Lett. 64, 1021 (1994), the content of which is incorporated herein by reference). An ECR plasma source potentially suitable for epitaxial growth by plasma enhanced CVD on large area substrates has been described, for example, in U.S. Pat. No. 5,580,420 to Katsuya Watanabe et al., the content of which is incorporated herein by reference. In industrial semiconductor processing, large ECR sources are, however, used for etching rather than epitaxy. Very high etch rates have been achieved in the case of III-V nitrides (see for example, Vartuli et al., Appl. Phys. Lett. 69, 1426 (1996), the content of which is incorporated herein by reference).
Yet other sources of high-density, low-energy plasmas are inductively coupled plasma ICP) sources. These sources have a number of advantages over ECR sources, such as easier scaling to large wafer diameters and lower costs. For a review of the different kinds of ICP sources, see Hopwood, Plasma Sources Sci. Technol. 1, 109 (1992), the content of which is incorporated herein by reference. The most common variants used for plasma processing are helical inductive couplers where a coil is wound around the plasma vessel (for example, see Steinberg et al., U.S. Pat. No. 4,368,092, the content of which is incorporated herein by reference), and spiral inductive couplers with flat coils in the form of a spiral (for example, see U.S. Pat. No. 4,948,458 to Ogle, the content of which is incorporated herein by reference). Plasma sources based on spiral couplers have the advantage of higher plasma uniformity, facilitating scaling to large substrate sizes (for example, see Collison et al., J. Vac. Sci. Technol. A 16, 100 (1998), the content of which is incorporated herein by reference).
While ICP sources normally are operated at a frequency of 13.56 MHz, operating at lower frequency has been shown to decrease capacitive coupling and thus leading to even lower ion energies (see U.S. Pat. No. 5,783,101 to Ma et al., the content of which is incorporated herein by reference).
Both, ECR sources and ICP sources are usually used for etching. Very high etch rates for GaN have been obtained also with ICP sources (see Shul et al., Appl. Phys. Lett. 69, 1119 (1996), the content of which is incorporated herein by reference). However, use of these sources for epitaxial growth of semiconductor quality materials is very rare. Recently, it has been suggested to apply an electrically-shielded ICP source for ion plating epitaxial deposition of silicon. This method has the obvious drawback of requiring a metallic collimator inside the deposition chamber (see U.S. Pat. No. 6,811,611 to Johnson, the content of which is incorporated herein by reference).
ICP sources can also be used for efficient cleaning of process chambers, such as chambers used for thermal CVD, where a remote plasma source is usually employed (see U.S. Pat. No. 5,788,799 to Steger, the content of which is incorporated herein by reference). Chamber cleaning is particularly important for semiconductor processing, where particulate contamination has to be kept as low as possible. Processing chambers equipped with a plasma source, such as an ICP source, do not of course require an additional remote source for efficient cleaning (see U.S. Pat. No. 6,992,011 to Nemoto et al., the content of which is incorporated herein by reference).
Whatever plasma source is used for generating a low-energy plasma for plasma enhanced chemical vapor deposition, when applied to III-V compound semiconductor growth, carbon incorporation into the growing layers is likely to occur to a much greater extent than in MOCVD.
Carbon incorporation results from the use of organic precursors in MOCVD and as suggested for LEPECVD (see U.S. Pat. No. 6,454,855 to von Känel et al. the content of which is incorporated herein by reference). The intense plasma used for cracking the precursors in LEPECVD is expected to greatly enhance unintentional carbon uptake, possibly to a degree unacceptable for device applications, since carbon acts as a dopant (see, for example, Green et al., J. Appl. Phys. 95, 8456 (2004), the content of which is incorporated herein by reference).
It is an objective of the present invention to avoid the drawbacks of prior art techniques mentioned above, such as carbon and hydrogen incorporation, high substrate temperatures, and low deposition rates. An additional major limitation of prior art techniques is a relatively small wafer size (two inch in production, up to 6 inch demonstrated for sapphire substrates). Increased scaling of silicon wafers of up to 300 mm (or more) is one of the objects of the present invention.